CHI@6
APRIt. 1 3 - Z a ,
~996
The Influence of Muscle Groups on Performance of Multiple
Degree-of-Freedom Input
S h u m i n Z h a i 1,2
Paul Milgram ~
William Buxton 2
~Department of Industrial Engineering and 2Department of Computer Science
University of Toronto, Toronto, Ontario, Canada, MS4 1A4
+1-416-978-1976
[email protected]
+1-416-978-3662
[email protected]
ABSTRACT
The literature has long suggested that the design of
computer input devices should make use of the fine,
smaller muscle groups and joints in the fingers, since they
are richly represented in the human motor and sensory
cortex and they have higher information processing
bandwidth than other body parts. This hypothesis,
however, has not been conclusively verified with empirical
research. The present work studied such a hypothesis in the
context of designing 6 degree-of-freedom (DOF) input
devices. The work attempts to address both a practical need
- designing efficient 6 DOF input devices - and the
theoretical issue of muscle group differences in input
control. Two alternative 6 DOF input devices, one
including and the other excluding the fingers from the 6
DOF manipulation, were designed and tested in a 3D
object docking experiment. Users' task completion times
were significantly shorter with the device that utilised the
fingers. The results of this study strongly suggest that the
shape and size of future input device designs should
constitute affordances that invite finger participation in
input control.
Keywords
Input devices, 3-D interface, 6 DOF input, motor control,
muscle group differences, hand, fingers, arm, homunculus
model.
INTRODUCTION
This paper studies the effects of using different muscle
groups in 6 degree-of-freedom (DOF) manipulation. In
particular, it investigates human performance differences in
6 DOF input control with and without the involvement of
the small muscle groups and joints in the user's fingers.
Increasingly, the user interfaces that we are designing and
using involve higher degrees of freedom than were found in
first generation GUIs. This places a higher load on the
operators of such systems. Consequently, it is all the more
important that the motor and cognitive resources of the
operator be used to their greatest effect. Many studies have
Permission to make digital/hard copies of all or parl of this material for
personal or classroom use is granted without fee provided that the copies
are not made or distributed for profit or commercial advantage, the copyright notice, the title of the publication and its date appear, and notice is
given that copyright is by permission of the ACM, Inc. To copy otherwise,
to republish, to post on servers or to redistribute to lists, requires specific
permission and/or fee.
CHI 96 Vancouver, BC Canada
© 1996 ACM 0-89791-777-4/96/04..$3.50
308
+1-416-978-1961
[email protected]
already been conducted which address issues in multiple
DOF input (e.g. [11], [12], [14], [25], [26] and [27]).
The study reported in this paper adds to this literature. It
investigates how human performance in 6 DOF tasks varies
according to the muscle groups employed. The
implications of this research are very practical. If
performance for a given task is higher when a particular
muscle group is employed, then future input devices can be
tailored accordingly, with affordances [8] that encourage the
use of that group.
The present study was inspired by Card, Mackinlay and
Robertson's "morphological analysis" of the design space
of input devices, which suggested "a promising direction
for developing a device to beat the mouse by using the
bandwidth of the fingers" [2]. This prediction was based
upon the well-established "homunculus model" from
neurophysiology (Figure 1) and some motor bandwidth
studies using Fitts' law tasks. However, empirical HCI
studies that support such a prediction have not been
conclusive. One candidate device which affords finger based
manipulation and could therefore conceivably "beat" the
mouse is a pen shaped 2 DOF input device such as the
stylus studied by MacKenzie, Sellen and Buxton [19]. In
their comparison of a stylus with a mouse, however, the
results were inconclusive: the stylus outperformed the
mouse in dragging but not in pointing tasks. Furthermore,
in the context of conventional GUIs, even if the stylus were
shown to result in higher performance, the issue may well
be moot. This is because the time to acquire the device (i.e.
the time for moving between the keyboard and pointing
device, which is much longer with the stylus) may
dominate the overall performance. We feel that such "side
effects" and the collective user experience with the mouse
as the "standard" 2 DOF computer input device may make
the replacement of the mouse with a finger operated device
a futile pursuit.
In the present study, for several reasons, we have chosen to
focus on 6 DOF tasks. First of all, there is not yet an
accepted standard 6 DOF input device. Secondly, few would
disagree that 6 DOF input tasks are much more demanding
than 2 DOF input tasks. Consequently, if one particular
design factor results in a slight advantage in a 2 DOF input
task, that same subtle advantage may manifest itself to a
much larger extent in 6 DOF input tasks.
APRIL t 3 - 1 8 ,
1996 ~ " ~
~
,..;:z;'o,:.°o,..
-.,,ow,°o
U
Figure 1 Homunculus model of somatosensory (left) and mot'or (right) cortex (Adapted from [23], G. H . Sage, Introductionto
MotorBehavior,©1977 Addison-WesleyPublishingCompanyInc. Reprintedby permissionof Addison-WesleyPublishingCompany,Inc.)
THE LITERATURE
As mentioned earlier, neurophysiological studies have
shown that various parts of the human body are
anatomically reflected in the brain disproportionately
relative to their physical size and mass, as illustrated by the
homunculus model (Figure 1). Of particular interest to the
present study is the fact that representations of the fingers
and the hands in both the somatosensory cortex and the
motor cortex are much richer than those of the wrists,
elbows and shoulders. We may therefore expect
performance enhancement if fine muscle groups (i.e.
fingers) are allowed to take part in handling an input
device. On the other hand, one should note that the
relationship between the size of cortical area and dexterity
has not been definitively proven in the field of
neuroscience.
One of the first studies on the effects of different body parts
in manual control was done by Gibbs [6]. In a one
dimensional target acquisition task, Gibbs studied task
performance of three different body pans: the thumb
(activated by the carpometacarpal joint), the hand (activated
by the wrist), and the forearm (activated by the elbow), in
both position and rate control systems with various control
gains and time delays. The performance ranking in Gibbs'
study was, in decreasing order: hand, forearm, thumb.
Hammerton and Tickner later replicated Gibbs' study in a 2
DOF target acquisition task [9]. Although Gibbs and
Hammerton et al subsequently argued about their
experimental methodology [7][10], the two studies in fact
arrived at very similar conclusions, that performance with
the hand (wrist movement) was 'superior to that of the
thumb and the forearm. This advantage was greater for
more difficult tasks, such as those with long time delays
[9]. Note that both studies found that the wrist was more
effective than the thumb. Neither Gibbs nor Hammerton
and Ticker included fingers in their studies, however.
The motor performance of different limbs has also been
investigated in various Fitts' law studies. Fitts' law
established the simple linear relationship: MT = a + b ID
in repetitive tapping tasks, where MT is the movement
time, ID = log2(2A/W) is the Index of Difficulty, A is the
movement amplitude and W is the width of the target area
[4]. The slope parameter b, in units of seconds/bit, is the
inverse of the motor system information processing rate.
Fitts' law studies have typically found this rate (l/b) to be
in the vicinity of 10 bits/second when the arm is involved
in the movement.
Fitts speculated that other limbs, such as fingers, may
show different processing rates [4]. Subsequent studies in
fact supported this hypothesis. Langolf, Chaffin, and
Foulke investigated the Fitts' law relationship using
amplitudes of A = 0.25 cm, A = 1.27 cm and A > 5.08 cm
[16]. For the first two amplitudes, the experiment was
carried out using a microscope. For the large amplitude
(>5.08 cm), the experiment was carried with direct vision.
Langolf and colleagues observed that for A = 0.25 cm
subjects moved the stylus tip (a 1.1 mm peg) primarily
with finger flexion and extension. For A = 1.27 c m ,
flexion and extension of both wrist and fingers occurred.
For A > 5.08 cm, the forearm and upper arm were involved
in the movements. On the basis of this method of
allocating actuation to different muscle groups by
controlling the range of movement, Langolf and colleagues
concluded that the information processing rates (l/b) for the
fingers, wrist, and arm were 38 bits/see, 23 bits/see and 10
bits/see respectively (see Figure 6 in [16]). This study has
been widely cited in the literature (e.g. [1]; [15]; [2]) as
evidence that fingers are among the most dextrous organs.
Card and colleagues recently reviewed Fitts' law studies
with various body parts (finger, wrist, arm, neck) and
pointed out the limitations of the widely used computer
input device - the mouse [2].
In summary, both neurophysiological studies (the
homunculus model) and Fitts' law studies suggest that use
of the small muscle groups (fingers and thumbs) should
result in better performance than the large muscle groups
(arm and shoulder). However some studies in manual
control, such as Gibbs' study [6] and Hammerton and
Tickner's study [9], are not completely consistent with such
a prediction.
Due to the theoretical motivation, most studies in the
literature tend to compare performance of different muscle
groups against each other. From a practical and ecological
309
C~'H 9 8
point of view, such a contrast is not necessary for the
design of 6 DOF input devices. The human upper limb as a
whole (from shoulder to finger tips) has evolved to be
highly dextrous yet powerful. Every part of it has its
purpose and function. What is needed in input device
design is to make use of all parts of the associated limb,
according to their respective advantages. The larger muscle
groups that operate the wrist, elbow, and shoulder have
more power and a larger range of movement. The smaller
muscle groups that operate the fingers and thumb have
more dexterity. When all the parts work in synergy,
movement range and dexterity can both be maximised.
TWO ALTERNATIVE DESIGNS
The preceding discussion suggests that performance
improvement in 6 DOF input does not necessarily lie
simply in moving operations from the large muscle groups
to the smaller ones, but rather in using the small muscle
groups in addition to the large ones. Motivated by this
hypothesis, we designed and implemented two alternative 6
DOF devices for this experiment: the Glove and the
FingerBall. The two designs were based on a single
common sensing technology, an Ascension Bird TM, and
both were free moving, isotonic devices operating in
position control mode [27]. The critical difference between
the two devices, therefore, lay in the involvement of fingers
in the manipulation of the 6 degrees of freedom.
APRIL
13-I8,
t996
The T-button was an essential component of the input
glove. Since the Glove requires rotation to be made with
the wrist, the elbow and the shoulder, its range of rotation
is limited. Whenever a limit is reached, the user needs the
clutch to disengage the manipulated object (by releasing the
T-button under the fingers) and restore the hand to a more
comfortable posture, and then recommence the
manipulation (by re-closing the fingers). This is very
similar to lifting a 2 DOF mouse and starting from a new
position on the mouse pad. We refer to this process as "reclutching".
The majority of existing designs of freely moving 6 DOF
devices, such as the "Bat" [25], the "Cricket" [3] and the
3D mouse [17] are similar to the glove design in assigning
wrist, elbow and shoulder muscle groups for manipulating
the six degrees of freedom; however, none of these make
use of the fingers.
The FingerBall
The second design is shown in Figure 3. This device has
been dubbed the FingerBall, to reflect the ability to operate
it with both arm and fingers.
The Glove
The first design was an instrumented glove (Figure 2) that
had been used in previous experiments [e.g. 28]. An
Ascension Bird TM magnetic tracker was attached to the
centre of the palm of the glove, the rotational centre of the
hand. Also mounted on the palm of the glove was a button
with a T-bar. The clutch could easily be pressed down by
closing the fingers. The entire glove device weighs 70g.
Figure 3 The FingerBall
Figure 2 The 6 DOF glove
The Glove design resembles many of the common virtual
reality input devices. When using the Glove, all translation
and rotation operations are carried out by the user's
shoulder, elbow and wrist, i.e., the gross joints and muscle
groups in the human limb. Other than pressing the Tbutton, the smaller, finer joints and muscle groups on the
fingers were not utilised for the 6 DOF manipulation.
310
The ball shape was chosen because a symmetrical ball
shape can easily be grasped, and manipulated by the fingers
in all directions. The FingerBall is designed to be held and
moved (rolled) by the fingers, wrist, elbow and shoulder,
in postures that have been classified as "precision grasp", as
opposed to "power grasp" [ 18]. Precision grasping, while
holding objects with the finger tips, places emphasis on
dexterity and sensitivity. In contrast, power grasping, while
holding objects against the palm, puts emphasis on
security and power. The FingerBall has been designed with
a versatile shape that is compatible with a variety of virtual
object shapes. This approach of choosing a versatile shape
is an alternative to the approach of making "props" that are
designed to resemble features of virtual objects being
manipulated [11]. The FingerBall instead is similar to the
concept of "bricks" [5], which are universal physical
handles to various virtual objects.
To take maximum advantage of finger operations, two
additional features would be desirable. One is to make the
ball tetherless, so that the user can roll it freely between her
APRIL 13-18, 1996 C~"~U96
fingers without interference. The second desirable feature is
that the ball be made of an elastic, conductive material, so
that the entire ball functions as a button that can be
squeezed from any direction. Since enabling technology for
wireless design is not easily available, the FingerBall
currently uses the Ascension Bird T M tracker mounted in the
centre of a sponge filled ball 6 cm in diameter. The entire
FingerBall weighs 66g. The cord of the Bird is pointed
away fi'om the hand in the null position, so as to maximise
the range of rotation without interference from the cord
(Figure 3). Furthermore, since the FingerBall can be rotated
up to 180 degrees in any direction, it did not require a
clutch as in the glove design for the task used in the
following experiment. It was therefore not necessary to
implement a button for the FingerBall in the experiment.
The 3Ball TM, manufactured by Polhemus [20], is a
commercial product similar to the FingerBall design. The
limitation of the 3Ball is its fixed button location. In order
to access this button, users can not freely roll the ball
between their fingers. Another spherical implementation of
a 6 DOF tracker, the "Cue Ball," was demonstrated by Dan
Venolia at the CHI'90 Interactive Experience.
EXPERIMENT
Experimental Set-up
Experimental Platform and Display
The experiment was conducted with a desktop stereoscopic
virtual environment, MITS (Manipulation In Three Space).
As illustrated in Figure 4, MITS consists of a SGI IRIS
4D/310 GTX graphics workstation, CrystalEyes T M
stereoscopic glasses, several 6 DOF input devices and a
software system developed by the first author.
Since the overall objective of this experiment was to
evaluate 6 DOF input interfaces, the emphasis in designing
the display was to provide the largest possible number of
3D spatial cues. This was to ensure that the task
performance was driven predominantly by differences in
input controller conditions rather than by difficulties in
perceiving depth information. A 120 Hz sequential
switching stereoscopic display was employed, which has
been shown to be a necessary feature fbr this kind of
experiment. To enhance the 3D effect, perspective
projection and interposition cues were also implemented.
The tetrahedra were drawn in wireframe so that all edges
and corners of the objects could be perceived
simultaneously. Subjects were asked to sit on a chair
approximately 60 cm away from the computer screen for all
experimental conditions (Figure 4).
Experimental Conditions
Two experimental conditions, the Glove and the
FingerBall, were used in this experiment. A pilot study
showed that the best per~brmances for both conditions were
achieved when the control display ratio (control gain) was 1
(both translation and rotation). In such cases, subjects can
take the advantage of the direct, one-to-one correspondence
between the input and the display.
Experimental 7ask
A 6 DOF docking task, illustrated in Figure 5, was used for
this experiment. This represents a common elemental task
that is involved in many higher level interactions. In the
experiment, subjects were asked to move a 3D cursor as
quickly as possible to align it with a 3D target. The cursor
and the target were two tetrahedra of equal size (4.2 cm
from the centre to each vertex). The edges and vertex
markers (bars and spherical stars) of both tetrahedra were
coloured s o that there was only one correct match in
orientation. The markers superimposed on each comer of
the tetrahedra served multiple purposes. The stars on the
target indicated the acceptable 3D error tolerance for each
vertex (0.84 cm). The two types of markers (stars and bars)
served also to differentiate the target from the cursor.
At the beginning of each experimental trial the cursor
appeared in the centre of the 3D space while the target
randomly appeared in one of five pre-set locations (14 cm,
11.5 cm, 12 cm, 7 cm, and 10 cm away from the centre of
the cursor in five arbitrary directions) and orientations
(l 10 °, 98 °, 106 °, 140 °, and 115 °, mismatched with the
cursor about five arbitrary axes). In contrast to the data
•
t
i~1~
urs°r
¢
Figure 4 Experiment Set-up
Figure 5 The Experimental Task
311
i,
C~'~,~ 9 6
gathering trials, during practice sessions the target appeared
in completely random locations and orientations.
15
t~
11
.~
10
Experimental Design
z
Subjects
Twelve paid volunteers who had no previous experience
with 6 DOF input devices were recruited. Two of them
failed to pass the screening test due to weak stereopsis
(using a Baush and Lomb Orthorator). The remaining 10
participated in the complete experiment. Their ages ranged
from 22 to 33, with a median of 29. Eight of the subjects
were right handed and two were left handed. Subjects were
asked to use their dominant hand with both input devices
(FingerBall or Glove).
Experimental Results and Discussion
First Analysis of the Overall Results
In the following data analyses, statistical model residuals
were analysed first and it was found that the residual
distributions were skewed towards lower scores. This is
typical when completion time is used as a performance
measure. Data with such skewed distributions do not
strictly meet the assumption of analysis of variance
(ANOVA). We therefore applied a common correction
technique, log transformation [13], to all time data in the
following ANOVA significance tests. For ease o f
comprehension, however, all numbers and figures
illustrating results are still presented according to the
original, untransformed scale.
13-t8,
1996
(sec)
During the trial, whenever a comer of the cursor entered
into the tolerance volume surrounding the corresponding
corner of the target, the star on that corner changed its
colour as an indication o f capture. Whenever all four
corresponding comers stayed concurrently matched for 0.7
seconds, the trial was deemed completed. At the end of
each trial, the trial completion time was printed on the
screen. The beginning of each trial was signalled with a
long auditory beep and the end of each trial was signalled
with a short beep.
A within-subjects design was used in this experiment, in
consideration of efficiency. Each subject was tested with
both of the two conditions, Glove and FingerBall, on the
same day. According to the results of our previous research
in [27-29], users' performance with 6 DOF isotonic position
control inputs tended to stabilise after 20 minutes of
practice. In this experiment, each condition was given
about 25 minutes of exposure, which comprised a short
demonstration, two warm-up trials, five tests and some
practice trials between tests. Each test consisted of two
identical blocks of trials. Each block had 5 trials with 5
distinctive initial target locations in random order. Test 1
started after a short demonstration and two warm-up trials.
Test 2, Test 3, Test 4 and Test 5 started 5, 10, 15, and 20
minutes after the beginning of Test 1 respectively. Subjects
were alternatively assigned to one of the two experiment
orders: Glove first (GB) and FingerBall first (BG). After
completing the first condition, each subject received a short
break before proceeding to the second condition.
312
E
i:
APRIL
~
'm Glove all
.~. 14
13
~5
~. 12
9
8
7
i
i
Test1
Test2
~
Test3
3'
1
Test4
Test5
Figure 6 Task completion times with FingerBall and
Glove (data from all 10 subjects)
Figure 6 shows the subjects' mean trial completion times
for each of the five tests. On average, task completion
times were clearly shorter for the FingerBall than for the
glove in each of the five tests. Repeated measure variance
analysis showed that overall performance scores for the two
devices were significantly different: F(1, 8) = 26.554, p <
0.005.
With both modes, subjects significantly improved their
performances over the course of the five experimental
phases: F(4, 32) = 34.04, p < 0.0001. In addition, the
performance differences between the two modes were
independent of experimental phase, as indicated by the
absence of a significant Device x Phase interaction: F(4, 32)
< 1, p > 0.5. The performance differences between the two
modes were also independent of initial target location /
orientation: F(4, 32) = 1.28, p = 0.3.
Other significant factors included Block: F(I,8) = 26.44, p
< 0.001. As stated earlier, each test consisted of two blocks
of trials. Completion times in the second block were
significantly shorter than for the first block, due to an
obvious learning effect.
The presentation order o f the two modes was not
significant: F(I,8) = 2.2, p = 0.17, but the Order x Device
interaction was significant: F(1,8) = 22.587, p < 0.005.
This could imply the presence of an asymmetrical skill
transfer as an artifact of the within-subjects design, a factor
which is an often overlooked and which can result in
misleading conclusions in such experimental research. As
Poulton [21, 22] argued, with a within-subjects design, the
actual skill transfer from one condition to another might
not be symmetrical, even though subjects' exposures to the
two conditions are ostensibly equalised.
In the present experiment, the Order x Device x Phase
interaction is also significant(F(4,32) = 8.7, p < 0.0001)
indicating that, if there was an asymmetrical transfer effect,
it varied with experimental phase. It is thus very likely that
asymmetrical transfer might have been a significant factor
in the early experimental phases but not in the later phases.
Two approaches were therefore taken, respectively described
in the following two subsections, to test if asymmetrical
rl!,
APRIL
13-18,
1996
~"U
~
skill transfer was likely to have been the only cause of the
performance difference between the FingerBall and the
Glove.
A Between-Subjects Analysis
In order to remove the possibility that the preceding results
were due solely to asymmetrical skill transfer, an
equivalent between subjects analysis was carried out using
only the data for the first device used by each group of
subjects, thereby eliminating any potential asymmetrical
ordering effects. In other words, subjects were divided into
two groups, where members of the FingerBall group were
the subjects who were tested with the FingerBall first and
the Glove later. Their data with the Glove were discarded
for the between subject analysis. Similarly, the FingerBall
data were discarded for the group who tested the Glove
first. This approach was expected to be much less sensitive
than the within subject analysis in the preceding section.
Figure 7 shows the results after discarding half of the data
in this fashion. Results of a repeated measure variance
analysis based on these between-subjects data were
consistent with the early analysis, i.e., completion times
with the FingerBall were significantly shorter than the
glove: F(1,8) = 3.6, p < 0.05. Learning apparently did not
reduce this difference, as the Phase x Device interaction was
not significant: F(4, 32) < 1, p > 0.5.
(see)
o~ 18
[] Glove
~ FingerBall
14
possible skill transfer effect between conditions did not
significantly affect performance in the final test.
On the basis of all three of the above analyses, we can
therefore confidently conclude that the FingerBall
significantly outperformed the Glove in the experiment.
The Effect of Clutching vs. Muscle Groups
From a practical point of view, the above analyses have
concluded that the FingerBall is a more efficient device
than the Glove. However, from a more theoretical point of
view, the cause of the performance differences is still not
clear. As described earlier, the FingerBall differs from the
glove in two major aspects: the use o f finger joints and the
absence of a button (and the accompanying re-clutching
processes). With the Glove, the re-clutching process takes
time to complete. Subjects usually carried out 1 to 3
clutches / dec[utches in each trial. This could therefore have
been the sole cause of the performance differences in the
above analyses, while obfuscating any effects of using
finger joints / muscle groups.
This issue had in fact been considered during the design
stage of the experiment, however, and the re-clutching
times (from the moment of disengaging to the moment of
re-engaging) accumulated in each trial were measured and
recorded during each trial. In the following analysis, the reclutching times were subtracted from the trial completion
times for the Glove condition. The resulting net score is
labelled as "C-R time". For the FingerBall condition, for
which no clutching was necessary, C-R time was identical
to the original completion time itself. Figure 8 shows the
mean completion time with the FingerBall, the mean
completion time with the Glove, and the re-clutching time
with the Glove, all from Test 5. The mean re-clutching
time in Test was 1.06 seconds, accounting for 10.7% of
the mean com }letion time with the Glove (Figure 8).
1 0 ",
tC
9 "
~"
"S"
.
time
8"
~a
Test1
Test2
Test3
Test4
7-
Test5
Figure 7 The FingerBall vs. the glove with 5 subjects in
each group (between-subjects)
:~
-
"
6"
5"
Performance Analysis of the Final Test
As an additional verification that asymmetrical skill
transfer effect was indeed not the fundamental cause of the
performance difference between the Glove and the
FingerBall, we also analysed users' performance in the final
test for each condition. As implied by the significant Order
x Device x Phase interaction in the initial within-subjects
analysis, it was very unlikely that asymmetrical skill
transfer was still in effect in Test 5, after 4 tests and 20
minutes o f practice with the second device. Indeed, a
repeated measure within-subjects variance analysis on the
data from Test 5 again confirmed the previous conclusions:
completion time was significantly shorter with the
FingerBall than with the Glove: F(1, 8) = 15.8, p < 0.005.
Furthermore, neither Order of presentation nor Order x
Device interaction was significant, implying that any
O
FingerBall
Glove
Figure 8 Mean completion time in Test 5
Note that the C-R time measure is biased against the
FingerBall condition, for two reasons. First, with the
FingerBall, the re-clutching process still exists effectively,
although not explicitly. From time to time subjects had to
move their fingers to different parts of the ball's surface to
make further rotation. This effort (and associated time) was
not taken into account by the adjusted C-R time measure,
since no explicit re-clutching time could be measured.
Second, during the re-clutching time w i t h the Glove,
subjects were not necessarily idle but probably were instead
313
engaged in mentally making decisions about what to do
next. Since it is known that mental rotation takes up a
certain amount of time [24], this time may overlap with the
re-clutching time in the Glove condition and is therefore
reduced in the C-R time measure. Nevertheless, C-R time
serves as a conservative measure to test if the use of fingers
really was advantageous. If the FingerBall still
outperformed the Glove, as measured by C-R time, the
advantage of using fine joints must therefore exist. The
converse may not be true, however.
Figure 9 shows the performance differences between the
FingerBall and the Glove as measured by C-R time. As can
be seen, the mean completion times with the FingerBall
were still shorter than the mean C-R times with the Glove.
Repeated measure variance analysis of C-R times collected
in Test 5 (final phase of the experiment) showed that the
difference between completion times with the FingerBall
and the C-R times with the Glove was still significant:
F(1,8) = 5.324, p < 0.05. Neither the order of presentation
nor its interaction with the device was statistically
sig,,ificant, suggesting that this difference was not caused
by asymmetrical skill transfer. This analysis therefore
further supports the conclusion that the use of different
muscle groups was indeed a major cause of the superior
performance of the FingerBall as compared to the Glove.
(see)
•
all
LIJ
9
Test1
Test2
i
•
Test3
Test4
Test5
Figure
9 Comparison between the FingerBal[ and the
Glove after discounting re-clutching time with the Glove
Subjective Evaluation
Upon completing the experimental trials, subjects
subjectively rated each of the devices on a continuous scale
ranging from -2 to +2 (-2: very low, -1: low, 0: OK, 1:
high, 2: very high). On average, the FingerBall received
higher ratings than the Glove (mean value: 0.78 vs. 0.60).
Of the 10 subjects, however, only 6 rated the FingerBall
higher than the Glove; the other 4 subjects rated the Glove
higher than the FingerBall (Figure 10). This is an
interesting contrast to the task performance measures. In the
final test, all subjects, except subject A, had shorter task
completion times with the FingerBall than with the Glove.
It appeared, therefore, that subjective preferences were
strongly affected by some salient features of the devices
other than performance. Subjects were encouraged to jot
down comments on features about which they felt strongly.
314
C~"~| 9 6
APRIL
B
E
F
£ubject
|3-18,
199&
High t
OK
Very
Low
A
C
O
J Glove
G
H
I
J
O FingerBall
Figure 10 Subjective Ratings of FingerBall vs. Glove
Upward arrows indicate that the FingerBall was preferred
Seven subjects felt that the cord with the FingerBall got in
the way. Three subjects did not like the wrist rotations
imposed by the Glove. Two subjects wrote that the
FingerBall was less natural than the Glove. Surprisingly,
one subject particularly liked the clutch function with the
Glove. One subject reported fatigue with both devices. The
fact that the Glove device is closer to what is typically
encountered in "virtual reality" systems could also have
been an influential factor in subjects' preferences.
CONCLUSIONS
On the basis of neurophysiological findings (the
homunculus model) and Fitts' law studies, researchers have
hypothesised that computer input devices that are
manipulated by fingers (the small muscle groups) should
have performance advantages over devices that are operated
by the wrist and/or elbow and/or shoulder (the large muscle
groups) [2]. It was believed that devices designed to
conform to such a hypothesis would outperform the mouse
- the standard 2 DOF computer input device. Follow- up
studies that test such a hypothesis or apply it to new
designs have not been reported in the literature, however.
Designing efficient 6 DOF input devices thus presents a
practical need, as well as a research opportunity to test the
hypothesis of muscle group differences. Through an
empirical study it was found that assignment of the muscle
groups in manipulating an input device was indeed a very
critical factor determining user performance. Our results
show that in a 6 DOF docking task, trial completion times
for an input device that included fingers during 6 DOF
manipulation (the FingerBall) were significantly shorter
than those of a device that excluded the fingers from the 6
DOF manipulation (the Glove). The results of our study
strongly suggest that future designers of such input devices
should design the affordances of input devices (i.e. shape
and size) such that the fingers are included in their
operation to whatever extent is feasible.
ACKNOWLEDGEMENTS
This paper is based on Chapter 4 of the first author's
doctoral thesis [27]. We would like to thank George
Fitzmaurice and other members of the ETC Lab and the
IRG group at the University of Toronto for their
comments, advice and assistance. Primary support for this
, ,,,,,
APRIL
13-~8,
1996
'~'~
~
work has come from the Information Technology Research
Centre (ITRC) of Ontario, the Defence and Civil Institute
of Environmental Medicine (DCIEM), and the Natural
Sciences and Engineering Research Council of Canada
(NSERC).
REFERENCES
1. Boff, K. R. and Lincoln, J. E. Engineering Data
Compendium - Human Perception and Performance.
Armstrong Aerospace Medical Research Laboratory,
Wright-Patterson Air Force Base, Ohio, 1988.
2. Card, S. K., Mackinlay, J. D., and Robertson, G. G. A
morphological analysis of the design space of input
devices. ACM Transactions on Information Systems, 9,
2 (1991), 99-122.
14. Jacob, R. J. K., Sibert, L. E., McFarlane, D. C., and
Muller, M. P. Integrality and separability of input
devices. ACM Transactions on Computer-Human
Interaction, 1, 1 (1994), 3-26.
15. Keele, S. W. Motor Control, in K. R. Boff, L.
Kaufman, & J. P. Thomas (Eds.), Handbook of
Perception and Human Performance. (pp. 30.1-30.60).
John Wiley & Sons, New York, 1986.
16. Langolf, G. D., Chaffin, D. B., and Foulke, J. A. An
investigation of Fitts' law using a wide range of
movement amplitudes. Journal of Motor Behavior, 8, 2
(1976), 113-128.
17. Logitech. 2D/6D Mouse technical reference manual. In
Fremont, CA 94555: Logitech Inc., 1991.
3. Digital Image Design Inc. The Cricket, product
information. New York, NY., 1993.
18. MacKenzie, C. L. and Iberall, T. The Grasping Hand.
North-Holand, Amsterdam, 11994.
4. Fitts, P. M. The information capacity of the human
motor system in controlling the amplitude of
movement. Journal of Experimental Psychology, 47,
(1954), 381-391.
19. MacKenzie, I. S., Sellen, A., and Buxton, W. A
comparison of input devices in elemental pointing and
dragging tasks, in Proc. of CHI'91: ACM Conference
on Human Factors in Computing Systems (New
Orleans, Lousiana, 1991), pp. 161-166.
5. Fitzmaurice, G. W., Ishii, H., and Buxton, W. Bricks:
Laying the foundation for graspable user interfaces, in
Proc. of CHI'95: ACM Conference on Human Factors
in Computing Systems (Denver, Colorado, USA, 1995),
pp. 442-449:
6. Gibbs, C. B. Controller design: Interactions of
controlling limbs, time-lags and gains in positional and
velocity systems. Ergonomics, 5, 2 (1962), 385-402.
7. Gibbs, C. B. Comments on 'An investigation into the
comparative suitability of forearm, hand and thumb
controls in acquisition tasks' by Hammerton and
Tickner. Ergonomics, 10, 4 (1967), 431-432.
8. Gibson, J. J. The ecological approach to visual
perception. Houghton Mifflin Company, Boston, 1979.
9. Hammerton, M. and Tickner, A. H. An investigation
into the comparative suitability of forearm, hand and
thumb controls in acquisition tasks. Ergonomics, 9, 2
(1966), 125-130.
10. Hammerton, M. and Tickner, A. H. Reply to
Comments by C.B. Gibbs. Ergonomics, 10, 4 (1967),
433.
11. Hinckley, K., Pausch, R., GoNe, J. C., and Kassell,
N. F. Passive real-world interface props for
neurosurgical visualization, in Proe. of CHI'94: ACM
conference on Human Factors in Computing Systems
(Boston, 1994), pp. 452-458.
12. Hinckley, K., Pausch, R., Goble, J. C., and Kassell,
N. F. A survey of design issues in spatial input, in
Proe. of ACM Conference on User Interface Software
and Technology (1994), pp. 213-222.
13. Howell, D. C. Statistical Methods for Psychology
(Third ed.). PWS-KENT Publishing Company, Boston,
1992.
20. Polhemus. 3Ball TM, product information. Polhemus,
Cochester, VT., 1993.
21. Poulton, E. C. Unwanted asymmetrical transfer effects
with balanced experimental designs. Psychological
Bulletin, 66, 1 (1966), 1-8.
22. Poulton, E. C. Bias in Quantifying Judgements.
Lawrence Erlbaum Associates, Hove and London, 1989.
!i!!ilil
23. Sage, G. H. Introduction to Motor Behavior, A
Neuropsychological Approach (2nd ed). AddisonWesley Publishing Company, 1977.
24. Shepard, R. N. and Metzter, J. Mental rotation of
three-dimensional objects. Science, 171, 3972 (1971),
701-703.
25. Ware, C. Using hand position for virtual object
placement. The Visual Computer, 6, (1990), 245-253.
26. Ware, C. and Jessom, D. R. Using the bat: a six
dimensional mouse for object placement. I E E E
Computer Graphics and Applications, 8, 6 (1988),
155-160.
! :!ii
i i~i!
27. Zhai, S. Human Performance in Six Degree of
Freedom Input Control. Ph.D. Thesis, University of
Toronto,1995. (http://vered.rose.toronto.edu/people
/shumin.html)
!il
28. Zhai, S. and Milgram, P. Human performance
evaluation of manipulation schemes in virtual
environments, in Proc. of VRAIS'93: the first IEEE
Virtual Reality Annual International Symposium
(Seattle, USA, 1993), pp. 155-161.
29. Zhai, S. and Milgram, P. Input Techniques for HCI in
3D Environments, in Proc. of Conference Companion
of CHI'94: ACM conference on Human Factors in
Computing Systems (Boston, 1994), pp. 85-86.
315